e768098d0e
Flake8 Lint / flake8 (push) Waiting to run
Spell check CI / Spell_Check (push) Waiting to run
tools_continuous_delivery / Private PyPI non-main branch release (push) Has been skipped
tools_continuous_delivery / Private PyPI main branch release (push) Failing after 2m42s
Publish Promptflow Doc / Build (push) Has been cancelled
Publish Promptflow Doc / Deploy (push) Has been cancelled
117 lines
4.5 KiB
Python
117 lines
4.5 KiB
Python
"""
|
|
EvalRunner — batch evaluation orchestrator for MAF workflows.
|
|
|
|
Bridges the gap between MAF's single-invocation workflow model and PromptFlow's
|
|
batch-level `aggregation: true` pattern.
|
|
|
|
Usage:
|
|
runner = EvalRunner(workflow, aggregate_fn, input_mapping={"values": "processed_results"})
|
|
result = await runner.run(dataset)
|
|
print(result.metrics)
|
|
"""
|
|
|
|
import asyncio
|
|
from dataclasses import dataclass, field
|
|
from typing import Any, Callable, Dict, List, Optional
|
|
|
|
|
|
@dataclass
|
|
class EvalResult:
|
|
"""Result of a batch evaluation run."""
|
|
|
|
per_row_outputs: List[Any]
|
|
metrics: Dict[str, Any]
|
|
errors: List[tuple] = field(default_factory=list)
|
|
|
|
|
|
class EvalRunner:
|
|
"""Runs a MAF workflow per row, collects outputs, then calls an aggregation function.
|
|
|
|
This mirrors PromptFlow's two-phase execution model:
|
|
Phase 1 — run each row through the workflow concurrently
|
|
Phase 2 — pass all collected outputs to the aggregation function
|
|
|
|
MAF workflows do not support concurrent execution on a single instance,
|
|
so `workflow_factory` creates a fresh workflow for each concurrent row.
|
|
|
|
:param workflow_factory: A zero-arg callable that returns a built MAF workflow.
|
|
:param aggregate_fn: A function that receives collected outputs and returns a metrics dict.
|
|
:param concurrency: Max concurrent workflow.run() calls (prevents rate-limit errors).
|
|
:param input_mapping: Optional rename map for transposed keys → aggregation function params.
|
|
For single-value outputs, _transpose produces {"values": [...]}. If the aggregation
|
|
function expects a different param name (e.g., "processed_results"), pass
|
|
{"values": "processed_results"}.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
workflow_factory: Callable[[], Any],
|
|
aggregate_fn: Callable[..., dict],
|
|
concurrency: int = 5,
|
|
input_mapping: Optional[Dict[str, str]] = None,
|
|
):
|
|
self._workflow_factory = workflow_factory
|
|
self._aggregate_fn = aggregate_fn
|
|
self._concurrency = concurrency
|
|
self._input_mapping = input_mapping
|
|
|
|
async def run(self, dataset: List[Any]) -> EvalResult:
|
|
"""Execute the full eval pipeline: per-row → collect → aggregate.
|
|
|
|
:param dataset: List of inputs to pass to workflow.run() (one per row).
|
|
:returns: EvalResult with per-row outputs, metrics, and any errors.
|
|
"""
|
|
semaphore = asyncio.Semaphore(self._concurrency)
|
|
per_row_outputs: List[Any] = [None] * len(dataset)
|
|
errors: List[tuple] = []
|
|
|
|
async def _run_row(index: int, row: Any) -> None:
|
|
async with semaphore:
|
|
wf = self._workflow_factory()
|
|
result = await wf.run(row)
|
|
per_row_outputs[index] = result.get_outputs()[0]
|
|
|
|
# Phase 1: run all rows concurrently (bounded by semaphore)
|
|
tasks = [_run_row(i, row) for i, row in enumerate(dataset)]
|
|
results = await asyncio.gather(*tasks, return_exceptions=True)
|
|
|
|
# Separate successes from failures
|
|
succeeded_outputs: List[Any] = []
|
|
for i, r in enumerate(results):
|
|
if isinstance(r, Exception):
|
|
errors.append((i, r))
|
|
else:
|
|
succeeded_outputs.append(per_row_outputs[i])
|
|
|
|
# Transpose outputs into aggregation inputs
|
|
aggregation_inputs = self._transpose(succeeded_outputs)
|
|
|
|
# Apply parameter name mapping if provided
|
|
if self._input_mapping:
|
|
aggregation_inputs = {
|
|
self._input_mapping.get(k, k): v for k, v in aggregation_inputs.items()
|
|
}
|
|
|
|
# Phase 2: call aggregation function
|
|
metrics = self._aggregate_fn(**aggregation_inputs)
|
|
|
|
return EvalResult(
|
|
per_row_outputs=succeeded_outputs,
|
|
metrics=metrics,
|
|
errors=errors,
|
|
)
|
|
|
|
@staticmethod
|
|
def _transpose(outputs: List[Any]) -> Dict[str, Any]:
|
|
"""Transpose per-row outputs into aggregation-ready keyword args.
|
|
|
|
- If outputs are plain values (str, int, float): {"values": [v1, v2, ...]}
|
|
- If outputs are dicts: {key: [row1[key], row2[key], ...]} for each key
|
|
"""
|
|
if not outputs:
|
|
return {"values": []}
|
|
if not isinstance(outputs[0], dict):
|
|
return {"values": outputs}
|
|
keys = outputs[0].keys()
|
|
return {k: [o[k] for o in outputs] for k in keys}
|